Decoding and Fusion in Sensor Networks with Noisy Observations and Communications

Abstract

In this paper, we study how to combine decoding and fusion at the access point (AP) in sensor networks for decentralized binary detection. We consider a scenario where all sensors make noisy observations of the same spatially constant binary phenomenon and communicate to the AP through noisy communication links. Simple distributed channel coding strategies are used, either using repetition coding at each sensor (i.e., multiple observations) or distributed systematic block channel coding. In all cases, the system performance is analyzed separating or joining the decoding and fusion operations. As expected, the schemes with joint decoding and fusion show a significant performance improvement with respect to that of schemes with separate decoding and fusion. Our results suggest that the use of multiple observations is often the winning choice at practical values of the probability of decision error at the AP.